Skip to content

Give your AI agent live data with MCP

AI agents5 min read

What MCP is, how to add the 1lookup.dev MCP server to a client, what its eight tools do, example prompts, and how to keep credit spend visible.

What MCP is

The Model Context Protocol, or MCP, is an open standard for connecting AI applications to outside tools and data. A server exposes tools, each with a name, a description and a defined set of inputs. A client built into an AI application finds those tools and lets the model call them during a conversation.

Without MCP, a model can only answer from what it learned in training and from what you paste into the chat. With it, the model can look up a phone number, check an email address or inspect an IP address at the moment you ask, and answer from the result. The client usually shows you each call and may ask for approval first.

Servers can run on your own machine or be hosted and reached over HTTP. A hosted server is the simpler choice for a data service, because there is nothing to install and every teammate connects to the same endpoint.

Adding the server to a client

The 1lookup.dev server is a hosted, remote MCP server at https://mcp.1lookup.dev/mcp. It works with Claude, ChatGPT, Cursor and other MCP clients. The steps differ slightly in each client, but the pattern is the same.

If the tools do not appear after you connect, start a new conversation, since many clients load the tool list when a chat begins. If a call is refused, reopen the sign-in step and confirm that the account you chose has credits left.

  • Open the settings area where your client manages connectors, tools or MCP servers.
  • Add a new remote server and paste the URL https://mcp.1lookup.dev/mcp.
  • Complete the sign-in step the client presents, so calls are tied to your 1lookup.dev account and credit balance.
  • Check that the client lists the eight tools, then start a new conversation so the model can see them.

The eight tools

Three tools cover the most common checks, and the rest handle everything else in the catalog, bulk work and your account.

The pairing to remember is search_products and run_lookup. When you do not know which product fits a task, the agent can search the catalog, read the credit cost, and then run the product it picked by slug. That gives you access to all 79 products through one connection.

  • validate_phone: validity, carrier, line type and risk for a phone number.
  • verify_email: deliverability, catch-all and disposable checks for an email address.
  • lookup_ip: location, ISP and VPN, proxy and Tor flags for an IP address.
  • run_lookup: run any product in the catalog by its slug.
  • search_products: find the right product and its credit cost for a task.
  • bulk_submit: start a bulk job from a list of inputs.
  • bulk_status: check a bulk job and fetch its results.
  • get_account: credit balance and recent usage.

Example prompts

Plain requests work, because the model picks the tool from the description. Being specific about what you want back helps.

Prompts also chain well. Ask the agent to validate a phone number and, only if the line type is mobile, check the carrier. Because each result comes back as structured data, the model can use one answer to decide the next call and explain its reasoning at the end.

  • Validate +1 512 555 0142 and tell me whether it can receive text messages.
  • Check jordan.avery@example.com and say whether it looks disposable or risky.
  • Look up 203.0.113.42 and tell me whether it is a VPN, proxy or datacenter address.
  • Which product tells me whether northwind.example has SPF, DKIM and DMARC set up, and what does it cost?
  • Run email-authentication-check for northwind.example and summarize what to fix first.
  • How many credits do I have left, and what did I spend most recently?

How credit spend stays visible

Tool calls that run a lookup spend the same credits as the equivalent API request, from the same balance. Results are cached for 7 days, so repeating an input in that window costs nothing, and requests that fail on our side are not charged.

Three habits keep spend predictable. Ask the agent to use search_products before an unfamiliar lookup, so the cost is stated before the call. Ask it to call get_account at the end of a session to report the balance and recent usage. And check usage in the dashboard when you want the full history.

Your plan includes a fixed number of credits each month, so spend has a ceiling that does not depend on the agent behaving well. Pay as you go adds extra credits only up to a monthly limit you set, and if you turn it off, requests return insufficient_credits until the plan renews or is upgraded.

Habits that keep agents safe to run

Treat bulk calls differently from single lookups. A bulk job can spend many credits in one step, so tell the agent to show you the number of inputs and the estimated cost, and to wait for your approval before it calls bulk_submit.

Keep the scope of each task narrow. An agent asked to enrich a list of three accounts is easy to check, while one asked to research the whole market can run many lookups you did not intend. Review the tool calls in the transcript, as you would review a script before running it on production data.

Remember what the data is. Results are inputs to a decision, and some are estimates, such as IP location. Ask the agent to report the fields it relied on, not only its conclusion.

Where MCP fits next to the API

Use MCP when a person is in the loop and the work is exploratory: checking a lead, investigating a signup, or answering a question about a domain. Use the API when a program calls a product on every event, since a fixed integration is easier to test and to rate limit.

Both draw from one credit balance, so you can prototype with an agent and then move the same lookup into code without changing how you pay.

[ Validate ][ Enrich ]
[Get started]

Run your first lookup in two minutes

Create a key, pick a plan and call any product in the catalog. Plans start at $29 a month.

[ Screen ][ Scrape ]